Machine Learning-Based Modeling of Ovarian Response and the Quantitative Evaluation of Comprehensive Impact Features

Clinical data-based modeling of IVF live birth outcome and its application

BACKGROUND

The low live birth rate and difficult decision-making of the in vitro fertilization (IVF) treatment regimen bring great trouble to patients and clinicians. Based on the retrospective clinical data of patients undergoing the IVF cycle, this study aims to establish classification models for predicting live birth outcome (LBO) with machine learning methods.

METHODS

The historical data of a total of 1405 patients undergoing IVF cycle were first collected and then analyzed by univariate and multivariate analysis. The statistically significant factors were identified and taken as input to build the artificial neural network (ANN) model and supporting vector machine (SVM) model for predicting the LBO. By comparing the model performance, the one with better results was selected as the final prediction model and applied in real clinical applications.

RESULTS

Univariate and multivariate analysis shows that 7 factors were closely related to the LBO (with P < 0.05): Age, ovarian sensitivity index (OSI), controlled ovarian stimulation (COS) treatment regimen, Gn starting dose, endometrial thickness on human chorionic gonadotrophin (HCG) day, Progesterone (P) value on HCG day, and embryo transfer strategy. By taking the 7 factors as input, the ANN-based and SVM-based LBO models were established, yielding good prediction performance. Compared with the ANN model, the SVM model performs much better and was selected as the final model for the LBO prediction. In real clinical applications, the proposed ANN-based LBO model can predict the LBO with good performance and recommend the embryo transfer strategy of potential good LBO.

CONCLUSIONS

The proposed model involving all essential IVF treatment factors can accurately predict LBO. It can provide objective and scientific assistance to clinicians for customizing the IVF treatment strategy like the embryo transfer strategy.

Harnessing Artificial Intelligence to Predict Ovarian Stimulation Outcomes in In Vitro Fertilization: Scoping Review

BACKGROUND

In the realm of in vitro fertilization (IVF), artificial intelligence (AI) models serve as invaluable tools for clinicians, offering predictive insights into ovarian stimulation outcomes. Predicting and understanding a patient's response to ovarian stimulation can help in personalizing doses of drugs, preventing adverse outcomes (eg, hyperstimulation), and improving the likelihood of successful fertilization and pregnancy. Given the pivotal role of accurate predictions in IVF procedures, it becomes important to investigate the landscape of AI models that are being used to predict the outcomes of ovarian stimulation.

OBJECTIVE

The objective of this review is to comprehensively examine the literature to explore the characteristics of AI models used for predicting ovarian stimulation outcomes in the context of IVF.

METHODS

A total of 6 electronic databases were searched for peer-reviewed literature published before August 2023, using the concepts of IVF and AI, along with their related terms. Records were independently screened by 2 reviewers against the eligibility criteria. The extracted data were then consolidated and presented through narrative synthesis.

RESULTS

Upon reviewing 1348 articles, 30 met the predetermined inclusion criteria. The literature primarily focused on the number of oocytes retrieved as the main predicted outcome. Microscopy images stood out as the primary ground truth reference. The reviewed studies also highlighted that the most frequently adopted stimulation protocol was the gonadotropin-releasing hormone (GnRH) antagonist. In terms of using trigger medication, human chorionic gonadotropin (hCG) was the most commonly selected option. Among the machine learning techniques, the favored choice was the support vector machine. As for the validation of AI algorithms, the hold-out cross-validation method was the most prevalent. The area under the curve was highlighted as the primary evaluation metric. The literature exhibited a wide variation in the number of features used for AI algorithm development, ranging from 2 to 28,054 features. Data were mostly sourced from patient demographics, followed by laboratory data, specifically hormonal levels. Notably, the vast majority of studies were restricted to a single infertility clinic and exclusively relied on nonpublic data sets.

CONCLUSIONS

These insights highlight an urgent need to diversify data sources and explore varied AI techniques for improved prediction accuracy and generalizability of AI models for the prediction of ovarian stimulation outcomes. Future research should prioritize multiclinic collaborations and consider leveraging public data sets, aiming for more precise AI-driven predictions that ultimately boost patient care and IVF success rates.

Analysis of the potential regulatory mechanisms of female and latent genital tuberculosis affecting ovarian reserve function using untargeted metabolomics

Female and latent genital tuberculosis (FGTB and LGTB) in young women may lead to infertility by damaging ovarian reserve function, but the regulatory mechanisms remain unclear. In this study, we investigated the effects of FGTB and LGTB on ovarian reserve function and potential regulatory mechanisms by untargeted metabolomics of follicular fluid, aiming to provide insights for the clinical management and treatment approaches for afflicted women. We recruited 19 patients with FGTB, 16 patients with LGTB, and 16 healthy women as a control group. Clinical data analysis revealed that both the FGTB and LGTB groups had significantly lower ovarian reserve marker levels compared to the control group, including lower anti-Müllerian hormone levels (FGTB: 0.82 [0.6, 1.1] μg/L; LGTB: 1.57 [1.3, 1.8] μg/L vs. control: 3.29 [2.9, 3.5] μg/L), reduced antral follicular counts (FGTB: 6 [5.5, 9.5]; LGTB: 10.5 [7, 12.3] vs. control: 17 [14.5, 18]), and fewer retrieved oocytes (FGTB: 3 [2, 5]; LGTB: 8 [4, 8.3] vs. control: 14.5 [11.5, 15.3]). Conversely, these groups exhibited higher ovarian response marker levels, such as longer gonadotropin treatment days (FGTB: 12 [10.5, 12.5]; LGTB: 11 [10.8, 11.3] vs. control: 10 [8.8, 10]) and increased gonadotropin dosage requirements (FGTB: 3300 [3075, 3637.5] U; LGTB: 3037.5 [2700, 3225] U vs. control: 2531.25 [2337.5, 2943.8] U). All comparisons were statistically significant at P < 0.05. The results suggested that FGTB and LGTB have adverse effects on ovarian reserve and response. Untargeted metabolomic analysis identified 92 and 80 differential metabolites in the control vs. FGTB and control vs. LGTB groups, respectively. Pathway enrichment analysis revealed significant alterations in metabolic pathways in the FGTB and LGTB groups compared to the control group (P < 0.05), with specific changes noted in galactose metabolism, biotin metabolism, steroid hormone biosynthesis, and nicotinate and nicotinamide metabolism in the FGTB group, and caffeine metabolism, primary bile acid biosynthesis, steroid hormone biosynthesis, and glycerophospholipid metabolism in the LGTB group. The analysis of metabolic levels has revealed the potential mechanisms by which FGTB and LGTB affect ovarian reserve function, namely through alterations in metabolic pathways. The study emphasizes the importance of comprehending the metabolic alterations associated with FGTB and LGTB, which is of considerable relevance for the clinical management and therapeutic approaches in afflicted women.

Predicting ovarian responses to the controlled ovarian hyperstimulation in elderly infertile women using clinical measurements and random forest regression

During the past decades, the number of elderly infertile women is obviously increasing in China, and more and more of them are likely to seek medical assisted reproductive technologies. As the in vitro fertilization/embryo transfer (IVF/ET) treatment presents special medical and psychological challenges to elderly infertile women, it is extremely helpful to perform the clinical evaluation and outcome prediction regarding IVF/ET outcomes. In this study, we retrospectively collected 12 clinical measurements in prior to the oocyte recovery for 689 elderly infertile patients (≥35 years of old), and used for predicting ovarian responses to the controlled ovarian hyperstimulation based on random forest regression models. Using different predictor sets and 10-fold cross validation approach, the Mean Square Error (±standard deviation) of prediction models varied from 7.56 ± 0.31 to 13.90 ± 0.37 in the training datasets, and the correlation coefficients between observed and predicted values ranged from 0.86 ± 0.02 to 0.72 ± 0.05 in the testing datasets. Among all clinical measurements involved in this study, the preovulatory follicle count (PFC), antral follicle count (AFC), and anti-Müllerian hormone (AMH) were revealed to be the most important features in prediction models. In conclusion, we successfully established the machine learning approach that could help the elderly infertile patients to better understand the most possible outcomes in subjecting to the controlled ovarian hyperstimulation.

Prediction model of gonadotropin starting dose and its clinical application in controlled ovarian stimulation

Background

Selecting an appropriate and personalized Gn starting dose (GSD) is an essential procedure for determining the quality and quantity of oocytes in the controlled ovarian stimulation (COS) process of the in-vitro fertilization (IVF) treatment cycle. The current approach for determining the GSD is mainly based on the experience of a clinician, lacking unified and scientific standards. This study aims to establish a prediction model of GSD, based on which good COS outcomes can be achieved with the influencing factors comprehensively evaluated quantitatively.

Material and methods

We collected a total of 1555 patients undergoing the first oocytes retrieving cycle and conducted correlation analysis to find the significant factors related to the GSD. Two GSD models are built based on two popular machine learning approaches, and the one with better model performance is selected as the final model. Finally, clinical application and validation were conducted to verify the effectiveness of the proposed model.

Results

(1) Age, duration of infertility, type of infertility, body mass index (BMI), antral follicle count (AFC), basal follicle stimulating hormone (bFSH), estradiol (E₂), luteinizing hormone (LH), anti-Müllerian hormone (AMH) and COS treatment regimen were closely related to the GSD (P < 0.05). (2) The selected model has good modeling performance in terms of both root mean square error (RMSE) (29.87 ~ 34.21) and regression coefficient R (0.947 ~ 0.953). (3) A comprehensive evaluation of influencing factors for GSD is conducted and shows that the top four most significant factors are age, AMH, AFC, and BMI. (4) The proposed GSD can approximate the actual value well in the clinical application, with the mean absolute error of only 11.26 units, and the recommended results can prompt the number of oocytes retrieved (NOR) close to the optimal number.

Conclusion

Modeling the GSD value with machine learning approaches is feasible and effective, and the proposed model has good clinical application for determining the GSD in the IVF treatment cycle.